Report

Self-Assembly of Proteins into Designed Networks

See allHide authors and affiliations

Science  03 Oct 2003:
Vol. 302, Issue 5642, pp. 106-109
DOI: 10.1126/science.1088074

Abstract

A C4-symmetric tetrameric aldolase was used to produce a quadratic network consisting of the enzyme as a rigid four-way connector and stiff streptavidin rods as spacers. Each aldolase subunit was furnished with a His6 tag for oriented binding to a planar surface and two tethered biotins for binding streptavidin in an oriented manner. The networks were improved by starting with composite units and also by binding to nickel–nitrilotriacetic acid–lipid monolayers. The mesh was adjustable in 5-nanometer increments. The production of a net with switchable mesh was initiated with the use of a calcium ion–containing β-helix spacer that denatured on calcium ion depletion.

The production of a designed arrangement of matter at the molecular level is a central goal of contemporary engineering endeavors (1). Besides micropositioning strategies (2, 3), the materials can also be placed by spontaneous self-assembly processes (4). Efficient biological self-assembly systems are, for instance, myosin filaments, the large heterocomplex ribosome, bacterial S layers (5), and membranes containing two-dimensional arrays of bacteriorhodopsin (6) or porins (7). Novel assemblies can be designed and produced with the use of engineered biological building blocks (5, 811). Molecular assemblies may be desiccated and then viewed by transmission electron microscopy or followed in situ with an atomic force microscope (12, 13). Here, we report a noncovalent planar network consisting of two biologically unrelated proteins and show how the mesh can be adjusted and also made switchable by varying the Ca2+ concentration.

We chose the structurally characterized proteins L-rhamnulose-1-phosphate aldolase (RhuA) and streptavidin as building blocks. RhuA is a C4-symmetric tetramer consisting of 274 amino acid residues per subunit (14). It resembles a flattened cube with dimensions of 7 nm by 7 nm by 5 nm. Streptavidin is a D2-symmetric tetramer with 159 residues and one biotin-binding site per subunit (15, 16). It forms a brick with dimensions of 6 nm by 5 nm by 4 nm and a pair of biotin-binding sites on each of the two 6-nm-by-4-nm faces (Fig. 1A). RhuA was fused with a His6 tag at each of the four C termini protruding from the top face, which was essential for the assembly on a lipid monolayer and for all purification steps.

Fig. 1.

Block construction and self-assembly in solution. All electron micrographs were negatively stained; all scale bars are 20 nm. The particles and nets were picked from numerous similar pictures. (A) View onto the flat 7-nm-by-7-nm top face (depth of 6 nm) of the C4-symmetric enzyme RhuA (14). The point mutations for biotin labeling are stated for one subunit. The C termini (Ct) at the top face carry His6 tags (18). Block bR is RhuA with eight tethered (dashed lines) biotins at the newly introduced cysteines, two of which are depicted as binding to a streptavidin (block S). The 6-nm-by-5-nm face (depth of 4 nm) of S is shown (15, 16). The bR·S unit is in turn bound to block bbS, which is a streptavidin with (the four depicted) bis-biotin linkers. (B) Block bR·S4 produced on a Ni-NTA column (18). Three samples and an average of 21 are given. (C) Block bR2·S7, which occasionally eluted from the Ni-NTA column, showing three samples and an average of 11. (D) Block bR4S12 produced as a by-product during self-assembly of bR and bR·S4. Three samples and an average of six are shown. The shortest distances between bR units are around 13 nm, confirming the presence of one S as spacer. (E) Networks produced by self-assembly of bR and bR·S4, reaching sizes of 50 nm by 50 nm.

In order to fasten two tethered biotins at each side face of RhuA at positions that would connect to juxtaposed biotin-binding sites of streptavidin, we introduced the mutations Asn133→Cys133, Lys261→Cys261, and Cys126→Ser126 (C126S) (17, 18). The C126S mutation was necessary for the removal of an interfering thiol. The cysteines were labeled with tethered biotin (18), giving rise to building block bR carrying eight biotins (Fig. 1A). The tethers were short enough to prevent streptavidin (S) binding on a corner so that bR acted as a rigid four-way connector, binding block S only at its side faces.

Equimolar mixtures of the building blocks bR and S in solution showed a strong tendency to form large globular aggregates. Therefore, we produced and purified the building block bR·S4, bR decorated with one S at each of its side faces (Fig. 1B). The planar bR·S4 then presents four pairs of biotin-binding sites that can be occupied by the tethered biotins of bR blocks. bR·S4 was prepared by rapidly mixing a solution of 2 μM bR with an eightfold molar excess of S. Rapid mixing was expected to reduce crosslinking into larger aggregates. Excess S and large aggregates were removed on a Ni–nitrilotriacetic acid (NTA) column. The eluate was analyzed by electron microscopy and revealed the expected bR·S4 block but also bR·S4·bR·S3 dimers and some higher aggregates (Fig. 1, B and C).

In a first self-assembly experiment, equimolar amounts of bR·S4 and bR were mixed in solution. Electron micrographs of the mixture revealed a collection of small planar quadratic nets where the distance between two bR molecules was 13 nm, which is in the expected range for a single streptavidin spacer (Fig. 1D). However, the network sizes did not exceed 50 nm by 50 nm, as shown in Fig. 1E, presumably because irregularities arose from insufficiently labeled bR (18).

In a second experiment, a monolayer of the lipid Ni-2-(bis-carboxymethyl-amino)-6-[2-(1,3)-di-O-oleyl-glyceroxy)-acetyl-amino] hexanoic acid (Ni-NTA-DOGA) (1820) diluted in dioleoyl phosphatidylcholine was spread on the surface of a solution containing building block bR·S4 (18). A small volume of an bR solution was then injected below the surface of the lipid film. After overnight incubation, the lipid monolayer with the interacting proteins was transferred to a carbon grid and negatively stained. The resulting electron micrographs showed patterns that resembled a woven fabric with the expected mesh size (Fig. 2). The network extended over more than 200 nm, in contrast to the 50-nm patches observed in the lipid-free approach, indicating that the planar monolayer indeed facilitated the formation of the planar network.

Fig. 2.

Representative negatively stained example of the networks formed by self-assembly of blocks bR and bR·S4 at a lipid monolayer. The lattices extend over more than 200 nm by 200 nm but contain irregularities. (Insert) A magnified sample.

The mesh was then extended by incubating the building block bR·S4 with bis-biotin-labeled streptavidin (bbS) produced as described (18). The result was analyzed by electron microscopy. Besides large disordered aggregates, we observed a collection of four-unit nets with an bR-to-bR distance of 20 nm (Fig. 3A), confirming the expected S·bbS·S spacer. For bR·S4 bound to a Ni-NTA column, the steps of loading with bbS, washing, loading with S, and washing can be repeated, resulting in the defined species bR·[S·(bbS·S)n]4 or bR·[(S·bbS)m]4 (where n and m are 0, 1, 2...) so that any uneven-numbered streptavidin spacer can be produced by mixing with bR or bR·S4, respectively. Even-numbered spacers are obtained if bR·[S·(bbS·S)n]4 blocks were bis-biotin-labeled as described for S (18) and then mixed with bR·S4. Accordingly, the design allows spacer lengths of any multiple of 5 nm.

Fig. 3.

Networks with streptavidin spacer rods produced in solution. The electron micrographs are negatively stained. The depicted samples are representative. (A) Block bR4·S16·bbS12 produced by self-assembly of R·S4 with bbS, showing three samples and an average of six. The distance between bR blocks is 21 nm (scale bar, 20 nm), in agreement with the expected S·bbS·S spacer. (B) Streptavidin rods of around five units produced by mixing bbS with a smaller amount of S (left). Longer rods of around 20 units obtained by mixing equal particle numbers of the five-unit rods and S (right). (C) Self-assembly of bR·S4 with (bbS·S)x bbS rods (18) of various lengths.

Furthermore, the mesh was increased by using less-well-defined streptavidin rods with bbS units at their ends. (bbS·S)x·bbS rods with x ranging around 2 (Fig. 3B) were produced by mixing bbS with a smaller amount of S. After removing excess bbS by washing in a Centricon (Millipore, Bedford, MA) (100 kD cutoff), the remaining rods were again incubated with a small amount of S, which resulted in a length multiplication (Fig. 3B). The limited curvature of these rods indicates their stiffness. In turn, stiffness and the scarcity of branching points show that the internal tether of the applied bis-biotin (18) was of an appropriate length. These rods may find application as stiff spacers between any biotin- or streptavidin-labeled compounds. With bbS units at their ends, the rods can connect bR·S4 blocks, forming large mesh networks. When bR·S4 was incubated with rod distributions of various lengths, irregular planar networks were observed (Fig. 3C).

In order to design a protein lattice with a switchable mesh, we selected a Ca2+-binding β-helix fragment of the enzyme serralysin from Serratia marcescens as a spacer (21). We assumed that this spacer would form a 2-nm solid rod with bound Ca2+ and would denature to a 20-nm mobile tether on Ca2+ removal by ethylenediaminetetraacetic acid (EDTA) (Fig. 4A). The fragment comprised three β-helix turns containing five times the characteristic motif GGxGxDxUx, where G is Gly, x is an arbitrary amino acid residue, D is Asp, and U is a large aliphatic residue. The first six residues of the motifs are involved in Ca2+ binding. In principle, this spacer can be infinitely extended by adding further repeats.

Fig. 4.

The 998-residue construct PGAL-β-PGAL (17, 18). (A) Ribbon model of the polypeptide indicating N and C termini of the construct. On Ca2+ removal, the β helix (scale bar, 2 nm) denatures, giving rise to a 20-nm peptide tether. (B) Electron micrographs of negatively stained solutions of PGAL-β-PGAL picked from numerous similar pictures (scale bars, 20 nm). The left side shows PGAL-β-PGAL after EDTA-free preparation (22), forming dumbbell-shaped particles (circled) corresponding to the model of (A). On the right side, the same preparation after treatment with 10 mM EDTA shows separated globular PGAL molecules (the denatured peptide is not visible).

The concept was tested with the 998-residue tandem fusion protein PGAL-β-PGAL consisting of the globular monomeric 468-residue protein 6-phospho-β-galactosidase (PGAL) (22) and the β helix (17, 18). The gene was expressed in Escherichia coli and purified as described (22). The product was analyzed by electron microscopy, showing dumbbell-shaped particles of the expected size that indicated that the spacer was short and presumably contained Ca2+ (Fig. 4B). After Ca2+ was removed with 10 mM EDTA, the electron micrographs showed pairs of particles at distances from 3 to 13 nm, which agrees with the expected distance distribution of random 20-nm coils (Fig. 4B). We conclude that EDTA had removed the Ca2+ ions and thus extended the spacer. Such a β-helix spacer may be used in future constructions, for instance by replacing the streptavidin linkers in the bR networks described above.

The designed planar networks of engineered proteins extend the two-dimensional supramolecular chemistry (23) to the realm of proteins. They follow extensive studies on DNA networks, making use of the recognition property of complementary single-stranded DNA (24). Compared with DNA, the proteins permit stiffer network elements and stronger associations. However, the strong binding of biotin to streptavidin (KD equals 10–15 M) prevents self-healing of irregular lattice points as occurs, for instance, during crystallization processes. This renders network assembly rather sensitive to inhomogeneities of the building blocks, which we experienced with respect to biotin-labeling of the cysteines of bR (18). Both problems may be ameliorated in future developments. Protein networks are well suited to placing functional proteins such as enzymes or membrane channels at defined relative positions, because such functions can be conveniently added by protein fusions (18). This may improve contemporary molecular lithography (25). Moreover, it permits the construction of spatially ordered biochemical teams, which may also include membrane components (26).

Supporting Online Material

www.sciencemag.org/cgi/content/full/302/5642/106/DC1

Materials and Methods

Figs. S1 to S3

References and Notes

References and Notes

View Abstract

Navigate This Article